Improving web accessibility for colour vision deficiency (CVD) users: A response time study

IMPROVING WEB ACCESSIBILITY FOR COLOUR VISION DEFICIENCY (CVD) USERS

A response time study

FÖRBÄTTRAD

WEBBTILLGÄNGLIGHET FÖR ANVÄNDARE MED DEFEKT FÄRGSEENDE

En studie med mätning i responstid

Bachelor Degree Project in Informatics 30 ECTS

Spring term 2018 Gunillah Edmark

Supervisor: Henrik Gustavsson

Examiner: Yacine Atif

Abstract

The digital world is acquiring more space in today’s society. With this being said, it is important to keep the web content interpretable for everyone, despite any kind of disability. This study focuses on colour vision deficiency and how to implement a real- time colour correction instantly in the web browser, without any additional assistive technologies needed. Two almost identical webpages have been developed for this project, with the difference being that one is with a colour correction function and one without, who then are measured in page-loading time to see if there is any greater loss of performance when executing the colour conversion.

Keywords: WCAG 2.0, Colour vision deficiency, Web accessibility, JavaScript, Colour

correction

Table of Contents

1 Introduction ... 1

2 Background ... 2

2.1 WCAG 2.0 ... 2

2.2 Colour vision deficiency ... 3

2.2.1 Anomalous trichromacy ... 4

2.2.2 Dichromacy ... 5

2.2.3 Monochromacy ... 5

2.3 Correction of colour vision deficiency ... 6

2.4 Specific corrections ... 6

2.4.1 Conversion from RGB to HSV ... 6

2.4.2 Conversion from HSV back to RGB ... 8

2.4.3 Extracting colour values from CSS3 file ... 8

3 Problem statement ... 9

3.1 Hypothesis ... 9

4 Method ... 10

4.1 Alternative methods ... 10

4.2 Ethics ... 11

5 Implementation ... 12

5.1 Literature Review ... 12

5.2 Prototype ... 13

5.3 Progression ... 14

5.4 Pilot Study ... 19

6 Evaluation ... 22

6.1 Presentation ... 23

6.2 Analysis ... 23

6.2.1 Measuring page-loading time of webpage ... 23

6.2.2 Measuring load time of colour correction script ... 26

6.2.3 Measurements of extra LOC required for colour correction ... 28

6.3 Conclusion ... 28

7 Concluding remarks ... 30

7.1 Summary ... 30

7.2 Discussion ... 30

7.3 Future work ... 31

References ... 32

1 Introduction

Companies and organizations place more of their information, services and communications through online channels while limiting accessibility beyond virtual reality. As a result of this development, the European Union (EU) is working on finding new ways to effectuate their goal of making the Internet more accessible also for people with different types of disabilities.

New union acquis (i.e. EU legislation) is based on Web Content Accessibility Guidelines (WCAG) 2.0, whose goal is to consolidate a single standard for web content and bind individuals, organizations and governments together internationally. With web content, they include elements such as text, images, sound and code that structures the information available online. There are several types of disabilities that perplex the use of information technology, such as mobility limitation, hearing- and learning disabilities or various types of visual impairment.

The focus in this thesis is colour vision deficiency, caused by the human vision system missing one or two of the cones in the retina of the eye or a shift in spectral sensitivity (Zhou et al.

2014). Colour vision deficiency (CVD), or colour blindness as it also can be referred to, is a disability that can generate limitations for some individuals when they go online, due to the reason that many developers and designers use colour to highlight important information, which can be hard to interpret for those who suffer from CVD. Correction of CVD can be difficult since CVD is split into several different conditions depending on the severity of the disease. It is also possible to suffer from multiple types of the disability simultaneously.

According to Zhou et al. (2014) is the use of only ‘web-safe’ colours together with a conflicting pair of colours, a reduction in the choice set and limits the web designer’s spectrum. A way around this issue is to replace the initial colours with colour that are perceptible and as similar as possible as the initial colour, which also maintains the colour diversity of the original webpage.

The general way to interpret web content that might contain conflicting colour choices is by using assistive technologies, these can be found through diverse applications or in the operating system of the device (Schmitt et al., 2012). However, these assistive technologies might need to be activated and this fact might not be acknowledged by the user. So, for this study, a simulated restaurant webpage was developed with a real-time colour correction in the web browser. On this restaurant webpage, a colour correction function was applied on top of the page in the menu bar, in the shape of a colour wheel. Activating the colour correction requires only a click for the user, in order for the colours to shift.

The colour correction function begins by converting the colour space from RGB to HSV since

HSV render the colours in a way that the human eye can better perceive them. HSV colour

space is also used for e.g. road traffic sign recognition and facial recognition (Saravanan et al.,

2016). After the colour space has been converted, a condition is set to make a shift in the colour

wheel if the colour set is conflicting. Lastly, the colour space is converted back to RGB from

HSV. The colour correction is executed in JavaScript and the webpage is built up with HTML5

and CSS3. The method used for this study is by performing experiments, measuring the time

interval when loading the restaurant webpage, with and without colour correction. The

hypothesis is that the colour and contrast correction will not elongate to the extent that the user

will exit the webpage.

2 Background

The EU has adopted new acquis regarding web accessibility, with the intention to make the web more accessible for everyone, regardless of any kind of disability. According to the EU (2017), web accessibility is not only about technical standards, web architecture or design but also about political will and moral obligations. This is comparable with the United Nations (UN) Convention for People with Disabilities (A/RES/61/106, 24 January 2007), where it is stated that the informative and communicative technologies need to be better adapted for the disabled - including the elderly, who tend to be inhibited/limited in their use of the Internet with non-adapted funds. In the European Commission’s proposal to accessibility requirements, alternative colour is suggested for conveying of information in e.g. general computer hardware and operating systems, self-service terminals, transport service webpages and mobile devices.

2.1 WCAG 2.0

WCAG 2.0 is a set of guidelines published by the World Wide Web Consortium (W3C)'s Web Accessibility Initiative (WAI), who develop standards for HTML, XML, CSS along with other protocols and languages. W3C consists of a mixture of organizations and individuals, who together have formulated a mutual standard regarding web content accessibility, in order to meet various needs that exist (Windriyani et al., 2014). The guidelines provide recommendations in the process of making web content more accessible for people with disabilities. Many developers do however not adapt to the guidelines during the process of their work. Kowto (2012) implies that many developers would be more eager to include these recommendations if they had been given easier access to knowledge and training in how to implement web accessibility to their design. Gilbertson et al. (2012) corroborate this notion, that it would be idealistic in order increase web accessibility, to offer education and spread more information about assistive technologies, based on several collected studies.

WCAG 2.0 was presented in 2008 and is the successor of WCAG 1.0 that was published in May 1999. When working with WCAG 1.0 it posed some problems determining whether or not a webpage met the set benchmarks, which in such cases caused difficulties for organisations that wanted to adopt this standard. The first edition was organised around guidelines that had checkpoints, whereas WCAG 2.0 instead is coordinated around four design principles of web accessibility, where each principle has guidelines and each guideline has a testable success criterion at level A, AA, or AAA (W3C, 2017). When a webpage fulfils level A criteria it means that the basic web accessibility features have been met and that the webpage does not e.g. solely use colour to indicate an action or to convey relevant information. Level AA, on the other hand, demands more in order to meet the requirements. An example of level AA criteria is that the webpage has to have a contrast ratio of 4.5:1 for background and foreground, while a level AAA requires a contrast ratio of at least 7:1.

Note that WCAG 2.0 does not give a step by step tutorial on how to implement web

accessibility, but rather what functionality is needed for the user. However, W3C does provide

a complementary techniques documentation in how to develop accessible web content that

meets WCAG 2.0 guidelines. Also available is another WCAG 2.0 document that tells you the

intent of each guideline and success criteria. Although a new version of WCAG is about to be

released later in 2018 due to the explosion of mobile web content. The new version is told to

be “backwards compatible” so earlier work with WCAG 2.0 will not be profitless.

There are multiple alternatives to validate a webpage to WCAG 2.0, Gilbertson et al. (2012) mention a tool called AChecker along with W3C’s own tools for a validation check. When testing a webpage for accessibility, with for an example AChecker, you start off by pasting the URL of the webpage and press “check it”. If there is no URL, there are other options for uploading the HTML file or pasting the HTML Markup. As can be seen in Figure 1, it is clearly described faults that are detected in the code and what is needed to be changed to pass the success criteria of WCAG 2.0.

Figure 1 Example of validation check with AChecker

2.2 Colour vision deficiency

The human eye’s colour vision consists of two kinds of photoreceptors cells contained in the retina, rod cells that are active in low light and cone cells that are active in daylight. A trichromatic (normal colour vision) observer obtains three kinds of cone cells, classified as long (L) controlling red, middle (M) controlling green, or short (S) controlling blue. If all three of those cones are fully functioning together with the light-sensitive pigments, you are able to see all magnitudes of colours (Figure 2) when the cones react correctly with the wavelengths of the light mixing the three colours (Zhou, Bensal & Zhang, 2014).

Figure 2 Colour vision for trichromats.

Approximately seven percent of males and less than one percent of females of European

descent have some form of inherited CVD (Zhou et al., 2014). It is most common that people

the X-chromosome. A male only has one X-chromosome in each of the 23 chromosome pairs, the other one in every pair being a Y-chromosome, while the female has two X-chromosomes, and inherited red-green CVD is a recessive trait. This is why men are more susceptible to suffer from CVD since both of the women’s X-chromosome need to be mutated while the male does not have the same “backup capability” due to the single X-chromosome (Poret et al. 2009).

Rarer is the blue-yellow CVD (tritanopia), which is inherited in a dominant pattern, meaning that one copy of a mutated gene in each cell is enough to cause a condition. Blue-yellow CVD is not associated specifically with the X-chromosome and therefore males and females are afflicted alike. The cause is a change in chromosome 7. This means that CVD that has not been inherited from parents is more likely to be the blue-yellow type of CVD. The first mentioned red-green CVD is caused by a change in the original X-chromosome (~ 23) which in turn implies that something went wrong in the meiosis following conception (Jorde et al., 2015).

Table 1 CVD types and incidence rates of Europeans (Jefferson & Harvey, 2006).

Medical Term Type Incidence

Anomalous trichromacy

Tritanomaly Deuteranomaly

Protanomaly

- 4,9%

1%

Dichromacy

Tritanopia Deuteranopia

Protanopia

0.002%

1.1%

1%

Monochromacy - 0.003%

Total - 8.005%

Colour-related vision problems have also shown to be dominant in groups of people with a non-diagnosed CVD, which proves that the colour correction would not only benefit the original crowd target but would also be appealing in general (Jefferson & Harvey, 2007).

There are several ways to test colour vision deficiency A standard procedure for detecting a red-green defect is with the Ishihara colour test. The creator of this test, Dr Sinobu Ishihara, developed this in 1917 at the University of Tokyo and the test is to recognise a number or an object in a varied dot pattern. A trichromat will be able to interpret the hidden object while a person with CVD will not (Poret et al., 2009).

2.2.1 Anomalous trichromacy

The most common kind of colour vision deficiency (CVD) is anomalous trichromacy (Nam et

al., 2005), where the three cones all exist, but one of them is impaired. There are three varieties

of colour reception depending on the damaged cone. Protanomaly refers to an L-cone

malfunction, deuteranomaly to an M-cone malfunction and tritanomaly where the S-cone is

affected. Protanomaly and deuteranomaly are the types of anomalous trichromacy that are most

common and both of these conditions means that persons have a hard time separating red from

green or green from red. This also affects the colours brown and orange and occasionally even

types of purple and blue hue. Tritanomaly is the most severe condition and is when the blue

cone is defected, causing trouble seeing blue and yellow colours.

Figure 3 Simulated colour vision for anomalous trichromats, from left to right:

protanomaly, deuteranomaly and tritanomaly.

2.2.2 Dichromacy

The condition when an entire cone is absent is called dichromacy which causes a two- dimensional colour space and as with anomalous trichromacy, it is also separated into three classes, protanopia (red), deuteranopia (green) and tritanopia (blue).

Figure 4 Simulated colour vision for dichromats, from left to right: protanopia, deuteranopia and tritanopia.

2.2.3 Monochromacy

Some may refer CVD as colour blindness, which can be deceptive based on the grounds that a complete lack of functioning cone mechanism, also known as monochromatic (or achromatopsia), is the severest form of an abnormal colour vision. (Jefferson & Harvey, 2006).

People who suffer from monochromacy have a complete lack of photoreceptors and are usually limited to a black-and-white world.

Figure 5 Simulated colour vision for monochromats.

2.3 Correction of colour vision deficiency

The focus of this thesis will be of correcting colours that are hard to interpret for the protanomalous and deuteranomalies, who are anomalous trichromats that have a deficiency on the red or the green cone in the retina of the eye. The main reason why these conditions are primarily being focused on is due to the fact that they are closely related to each other and have a similar recolouring procedure. Another reason is that they are the most common CVD disorders, as can be seen from Table 1, page 4. Yet, as Jefferson, et al. (2006) also stated, is that it is not only applying the right colour correction that is important while working against the WCAG guidelines, but also the contrast ratio as discussed in their article from 2006.

“Ensure that foreground and background color combinations provide sufficient contrast when viewed by someone having color deficits or when viewed on a black and white screen.”

(Jefferson et al., 2006, pp. 44)

An example of an impaired foreground and background combination would be red and green, as this duo, is the most common CVD deficiency and would not contrast well at all. This would mean that people who suffer from a red-green colour vision defect would not be able to interpret the information displayed on the screen. As proposed by Jefferson et al. (2006), this study will be divided into four distinct parts.

1. First by selecting a set of colours, e.g. pink, red and orange, which can be confused with greens by deuteranopes (Ribeiro et al., 2013). This is an important step due to the fact that optimization can be ineffective if the number of corrected colours are too many.

2. Secondly, find the right colour and brightness differences for these colours, once again referring to the importance of contrast ratio.

3. Later in the third step construct an optimization step that stores these target distances.

4. In the fourth and final part is to insert the target distance colours in to the document.

2.4 Specific corrections

To achieve the CVD colour correction, JavaScript and CSS3 will be used, together with the HTML5, in order to demonstrate and measure to see if the hypothesis is correct. As previously mentioned, earlier studies have mainly been focused on converting images on a pixel level (e.

g. Ribeiro et al., 2013), whilst this study will instead measure the response time when loading the page after the colour correction link has been pressed.

The main focus of this study will not be on the images, but instead on the text and background on the webpage. When collecting the measure values, benchmarking will be used.

2.4.1 Conversion from RGB to HSV

It is a well-known fact that many of the displaying devices use RGB model to encode colour (Ribeiro et al., 2013). RGB stands for red, green and blue and by merging these colours’

intensity the spectrum of colours is acquired. RGB is however not always the best choice, given

that it is dubious in how to change the values in order to get for example a darker red or a

lighter green and the RGB colour space also has its issues of representing shading results. As

an example, to get the colour black, RGB values are minimum at R=0, G=0 and B=0 and to

get white the numbers are instead R=255, G=255 and B=255. In order to easier specify colour

correction, HSV (hue, saturation, value) will be used because it is more sufficient for a

precision of colour and this model describes colours more exactly as to how the human eye perceives them (Saravanan et al., 2016).

In order to describe RGBàHSV conversion, HSV can briefly be explained as followed: H represents the colour and ranges between 0-360 degrees, S determines the colour purity and ranges between 0-100 percent. As such, a low percentage displays a greyish colour while a high percentage gives a deeper colour. Finally, V decides the brightness of a colour and ranges between 0-100 percent. A decreasing V equals an increasing blackness (Chernov et al., 2015).

So, the correction for this project will start by converting the coloured pixels from RGB to HSV as Ribeiro et al. (2013) and Ching et al. (2010) do in their paper. There are several optional algorithms for RGBàHSV conversion and the one which will be presented in this paper is the classic real-valued algorithm made by Alvy Ray Smith, who in 1978 created the HSV colour space (Chernov et al., 2015). The transformation algorithm of converting RGB to HSV starts off with finding the minimum (m) and maximum (M) of R, G and B (1).

(1) (2)

When maximum and minimum values have been collected, V (value) will be assigned M, then d (delta) will be calculated between M and m. In case that d is equivalent to 0, S need to be assigned with 0 and returned. Then, S need to be calculated as a ratio of d and M (2). Moving on to calculating the H value, is dependable of what M and m are (3).

(3)

In conclusion, the conversion concerns finding the maximum and minimum of RGB. When these are equivalent is when S is 0 and H is undefined and also V will unceasingly be equivalent to the maximum RGB value. To calculate the saturation, the maximum of R, G or B is subtracted from the minimum of R, G and B and then divided with the maximum of RGB component (4).

(4) (5)

Next step is determining the correct H which needs a fraction (5), where the “middle

component and a sector offset are determined according to previously found M and m” as

Chernov et al. describe this (2015, p. 332).

2.4.2 Conversion from HSV back to RGB

After the conversion from RGB to HSV, the conversion goes back to RGB from HSV. This can be done by then seeing if S is equivalent to 0, all RGB will be assigned to V, after that returned and finally multiply H by 6. Then the hue sector index (I) has to be established, this is done by taking the biggest preceding integer of H. Moving on, the hue fraction is determined by subtracting I from H and the values (M, N & K) of RGB components (6).

(6)

The last step in the backward conversion is to assign R, G and B with relevant values, matching to I (7).

(7) 2.4.3 Extracting colour values from CSS3 file

The RGBàHSV and HSVàRGB algorithms then need to be applied in a JavaScript function

in order to perform this colour correction. The reason why there is a need for a back-conversion

from HSVàRGB is that it is not possible to specify the CSS colour in HSV values. In the

JavaScript colour correcting function, it is also needed to extract the colour values from the

CSS3 file.

3 Problem statement

According to Reichling et al. (2013), it showed in a global analysis that 24.9% of European government portals and ministry websites failed and detected barriers to accessibility.

Although, the European tests did receive the best results compared to the rest of the testing areas, as for example Asia and Africa who had a failure percentage between 39 to 42. Yet merely within the EU member states are there over 80 million people with some kind of disability (Kowtko, 2012), concluding that a big market is excluded when developing web pages and mobile applications. Approximately one-quarter of all these people suffer from one of the types colour vision deficiency, hence the focus on CVD.

There are already solutions for Android and other operating systems that are implemented in their software, as mentioned before, you normally have to activate this mode by yourself (Schmitt et al., 2012). This is an action which is not always known by the user and is a reason why it is a good idea to move the solution into the web browser, which would make it more noticeable for the user. However, it is also important to keep the response time within what is acceptable for the user, as Hoxmeier et al. (2000) describes, that lowered user satisfaction is significant because it might lead to discontinued use, force the user to find alternative sources of information and results from their study showed that satisfaction does decrease as response time increases.

So, the question asked for this project is:

- How is the response time effected by a colour correction?

Similar research has been found about colour correction for CVD, but none specifically regarding measure of the response time between an implemented accessibility webpage and non-accessibility webpage, with JavaScript. Another interesting aspect to measure is the extra work that will be put in to produce the additional code that will transform the webpage into a colour vision corrected mode for CVD disables and this will be measured in lines of code (LOC).

3.1 Hypothesis

A null hypothesis and an alternative hypothesis are stated for this project. The null hypothesis claims that there will be no affect in response time with the colour correction function on the webpage. The alternative hypothesis is that the colour and contrast correction will not elongate to the extent that the user will exit the webpage. It has shown that e.g. high-resolution images take a longer time to translate as colour difference calculation depends on the overall number of pixels in an image (Ruminski et al., 2010), although the expectation is that this will not affect the response time excessively.

The ambition with counting the additional colour correction script in LOC is to prove that the

LOC will not contribute to an excessive workload that will cause it to be non-sustainable. This

study is primarily based on Saravanan et al. (2016)’s idea on converting the RGB colour space

to HSV colour space since this colour scheme is better for the human eye and therefore is an

important element of this colour correction procedure.

4 Method

For this project, a simulated webpage will be produced and thereafter implemented with colour vision deficiency correction where the efficiency will be measured in page-loading time in milliseconds to see possible loss of performance. Ruminski et al. (2010) too adopted a quantitative experiment where they collected an average time of colour transformation, however, their focus was on image processing methods. The guidelines of WCAG 2.0 will be used as a foundation for the colour and contrast correction, together with Reichling’s et al.

(2013) conclusions in how to implement the accessibility modifications in the most efficient way, with the aim to increase usability for the people with CVD. In order to facilitate the evaluation of hypothesis outcome, the aim has been split the simulation into lesser subjects, on a priority scale with 1 holding the highest rank:

1. Implement two different correction functions on RGB data.

2. Implement a traversal of documents with JavaScript and modification of CSS3.

3. Adjust colours in document with CSS3 in RGB.

The research methodology which will be used, is the experiment method, by reasoning that it is the most adequate choice for collecting and comparing the given results. The measurements will be compared to the original webpage’s response time to see eventual failure in the time interval. Several methodical challenges are anticipated. For instance, when measuring the time interval between the original webpage and the colour corrected version, there will be a natural delay due to the extra code needed for the adjustments. Another issue is regarding the network that the tests are run on, which also can affect the results.

Zhou et al. (2014) have a similar approach to their study where they applied six different colour themes to 14 randomly selected mobile websites and also focused on red-green colour vision deficiency. In contrary to this study, Zhou et al. chose websites in several categories such as blogging, corporate and education. The colour themes Zhou et al. applied were parsed with the CSS file of each website to extract colours and each colour was converted to luminance, chroma and hue angle values. However, their study is executed on mobile websites and evaluated by four factors that would influence a user´s experience in web browsing, and not by comparing response time between a colour corrected and non-colour corrected webpage. The four factors that they evaluated from are differentiability, naturalness, subjective response naturalness and subjective response differentiability. Differentiability measures the difference between the colours used together, naturalness measures the closeness between replacement colours, subjective response naturalness measures the preservation of subjective factors such as weight and activity of the replacement colours and subjective response differentiability measures the preservation of relative difference of subjectivity of replacement colours. Zhou et al. concluded that the experiment results from their proposed colour adaption method improves mobile web accessibility through their colour adaption method, therefore this study focuses on response time instead.

4.1 Alternative methods

There are generally three alternative strategies that can be used as a research methodology:

experiment, survey or case study (Wohlin et al., 2012). A survey is a technique which is mostly

used in social sciences and its purpose is to describe people’s behaviour, knowledge and

attitude (Wohlin et al., 2012). Since the goal of this project is to see the difference in response

time, a survey would be redundant since the outcome would not be of any use to answer the

hypothesis. If it, on the other hand, would have addressed if there is a need for colour vision deficiency correction or if the user’s capability to interpret the webpage after implementing the disability aid, then this would have been an ideal technique.

Case studies are the last of the three considered strategies discussed for this project, a strategy that according to Wohlin et al. (2012) has been seen as less valued by critics’, due to the fact that it does not provide the same statistical significant conclusion as, for an example, experiments do. However, a case study is told to provide a greater real-life context than an isolated experiment, due to the multiple sources of evidence (Wohlin et al. 2012). A case study would have been a good choice in a later stage and a good idea of use as a research methodology for further work within this subject.

4.2 Ethics

Since there are no human subjects involved, the ethical aspect concerning exposure of

confidential and/or sensitive data is non-existing (Wohlin et al., 2012). This will also apply to

the webpage that will be developed for this research and it will exclude any type of data that

could contain personal information. Also, all code, implementation, results, hardware

specifications, other resources, etc. will be presented as appendices and the code is also

available on GitHub in order to be replicated for further research and testing. The experiment

will not request the user to reveal their individual type of colour vision deficiency, as it is not

relevant for this study since the goal is to measure the response time between a corrected and

a non-corrected webpage. The correction is done in the web browser so there is no request for

the user to provide any information beforehand, the settings is also saved locally which means

that there is no information collected and sent further about the user’s choices made on the

webpage.

5 Implementation

This section describes the process of the implementation and adaption of the algorithms, the simulated webpage and structure of the study.

Initially what was needed to be created and executed to answer the hypothesis was:

- An HTML file structuring the webpage

- A CSS file describing the visual with colour and design

- A JavaScript file performing the colour conversion from RGBàHSV, HSVàRGB and code distributing which elements that execute the rotation of hue

5.1 Literature Review

The disadvantages of the RGB scheme are numerous, and as previously explained, it is not intuitive for human interpretation and chrominance and luminance properties are not divided as well as all the components are highly correlated altogether. As Chernov et al. (2015) describe is HSV a superior colour space compared to RGB according to earlier experimental results.

The work of Ribeiro et al. (2013) and Ching et al. (2010) will be two of the primary scientific articles used for this project. In line with the aim of Ribeiro et al. (2013) and Ching et al. (2010), the focal point of this baccalaureate thesis is on red and green colour vision deficiency, whereas in contrast to Ribeiro and Ching there will not be a focus on images but instead mainly on text as well as background colours. Incorrect colour combinations can be of concern for the colour vision deficient user that enters a webpage and could cause misconception when interpreting the information.

Kowto (2012) also concludes that the Internet might be considered simple for the average individual but for those with disabilities it can be challenging, especially when neither the web accessibility guidelines nor web assistive technologies have been taken into consideration during development. Ribeiro et al. (2013) are also consenting with the contrast issue and describes how it could be a complex task for the visually impaired to interpret these kinds of design choices. Also, as previously stated, Zhou et al. (2014) provided the most similar artefact approach when they applied different colour themes to mobile websites.

A related point to consider is that none of the reviewed literature contains colour correction with executing actions in JavaScript, hence the most code is influenced by examples found on developer community StackOverflow. A larger amount of the scientific articles found about colour vision deficiency and web technologies instead discusses the colour conversion of images, different types of correction algorithms and front-end web accessibility in general.

The conversions will be executed with JavaScript, an object-based language, which is

considered as the programming language of the web (Flanagan, 2011). Acquiring information

about colour correction in JavaScript textbooks resulted in a conflict due to the reason that not

much has been written about this subject. For example, Flanagan (2011) solely describes how

colour is used when styling the webpage and also different colour spaces are mentioned as for

example RGB as well as HSL (hue saturation, lightness), however, HSV colour space is never

mentioned. Regarding HSL, some might think by the name that it can be comparable with

HSV, but there are quite some important differences. For once, there is a difference in

definition of saturation (S) as well as between the L (lightness) and the V (value), wherein V

it is grasped as the amount of light of any colour while L only handles the amount of white.

This factor contributes to the reason why S is defined differently, due to the deviation of L and V it is scaled differently.

According to Hoxmeier et al. (2000) is a response time of 1-2 seconds proven to be satisfying to the user, but response times longer than 15 seconds are disruptive. Hence, the results from the collected measurements in this study will compare to these numbers.

5.2 Prototype

In order to measure the response time between a non-corrected webpage and a colour corrected webpage to see if a real-time implementation would be of value, a simulated restaurant site was fabricated for this paper where the time measurements are withdrawn from. JavaScript functions are included on the webpage that will collect the requested data needed for analysis.

The webpage consists of a navigation bar, a header containing several images, followed by three sections containing text, another image area and finally at the bottom, a footer with additional information about the simulated restaurant company. The correction of this webpage will, as previously mentioned, start by converting all text and background colour including the menu, headlines and other text sections. The webpage is structured with HTML5 and styled with CSS3, both of whom are markup languages. CSS is more specifically a style sheet language that tells how the webpage is going to look with e.g. colour and size of different elements in order to personalise the site, while the HTML creates the structure and informational content of the webpage.

When designing for accessibility it is crucial to make sure that this code structure follows the correct guidelines as W3C suggests in WCAG 2.0. This is necessary so that for example low vision users can override the existing CSS style sheets in order to make it possible for them to interpret the webpage with larger fonts or different colours. Another example is to remember to simply declare a language for the HTML code, so if the user also is using assistive technology, such as screen reader or text-to-speech, it won’t create a blockade when entering the webpage, this according to Gilbertson et al. (2012).

But as previously mentioned, not all users are aware of their own colour vision deficiency, which is the main subject of this paper and through this opens a market making it possible to enable colour corrections in the browser. To activate the colour conversion for the webpage made for this project, the user presses the button with a rainbow colour scheme. Besides HTML5 and CSS3, JavaScript is used to perform the actions as mentioned above.

The plan was to produce a webpage that would illustrate a real-life situation, in this case

imagining a colour vision deficient visiting a restaurant webpage. With the intention of looking

authentic, the webpage was given appropriate information with fitting images, that was

retrieved via Creative Commons. Creative Commons is a non-profit organisation, providing

free licenses that help to extend the creators possibility to share their content and making it

available to the public. The Creative Commons licenses are arranged into six standard CC

licences and the images that occur on the simulated restaurant webpage are all used with

copyrights given to the creators in an additional HTML-file on GitHub (32e33d9). Originally

the idea was to also include colour correction of the images on the webpage, however due to

time constraints this step was excluded.

Figure 6 Screenshot of restaurant webpage

Before developing the webpage, a mock-up was created, this mock-up was used as a template for code structure, as can be seen on GitHub (32e33d9). Some alteration was done during the process from original idea to result (407f572). When the webpage was designed, it was decided to also consider some of the other guidelines from WCAG 2.0 for web accessibility. For example: not using only colours when providing important information on the webpage, not overcrowding the webpage with too much information and colour combinations and use high contrast colours for text, buttons and links on the webpage.

5.3 Progression

The first HTML, CSS and JavaScript files that were uploaded on GitHub was sorted in a separate folder named “Webpage”. This was not used later since these files were only created to test a JavaScript body onload-function (Figure 7) that would transform the background- colour from the one originally set in the CSS-file (fe3aedd).

Figure 7 Code example of ”body onload”

After the first code had been tested, the work took off in a new direction, as mentioned earlier.

The original idea was of creating a simple standard business webpage, but it was then decided

to target a specific business area, hence the restaurant webpage (Figure 6). To provide some

explanation about the structure of the webpage, it was to put the menu on top of the page with

only a few menu bar items, in order to make the webpage not too challenging for the user to

interpret. When designing a webpage, it is also recommended, as by Sik-Lányi et al. (2017), to

use patterns or higher contrast in order to make the webpage generally more accessible and

making it easier for the user to construe the content. A large image area was put under the menu

section with the intention of giving the webpage a more attractive appearance. The image area contains a slideshow function that enables the user to click an arrow on both sides of the image to change the picture. When hovering over the arrows, the area transforms and turns the background area of the arrow black and highlights the possibility to click on it. In case the user would not be able to see the arrows, there are three dots underneath with a strong contrast colour against the dark background (Figure 8) to indicate this possibility.

Figure 8 Image slideshow

In the same area as the three dots, the user can get quick access to book a table at the restaurant.

This was implemented with the intention to once again provide a simple design that does not contain too much information and make it easy to use. The following section is representing an information area about the restaurant with just plain text, followed by a new image section that contains a link directing the user to the restaurant menu.

The next section before the footer of the webpage encloses contact information and a world map image, that disappears when the window is less than 700 pixels wide. This was accomplished with Media Queries, which is a CSS technique that renders content and adapts to the conditions that are set to different screen resolutions. Between the menu-section and the footer is a form that invites the user to sign up for a newsletter to easily receive information from the restaurant regarding new menus and special events.

Figure 9 Media Query properties for world map image

After the design template had been set, the actual HTML structure, CSS styling sheet and JavaScript events were created after the prototype, where the measurements tests of response time later are being performed at (aaa6a47). The elements that the colour correction is planned to be executed at are:

- General text colour for paragraph (p) tag standard - Text colour for headings (<h1> to <h6>)

- The background colour of the entire webpage

- Menu: background colour, text colour and text colour while hovering over a chosen alternative link

- Header: text colour

- Footer: background colour and text colour

These elements are the prioritised ones for this bachelor project and the objects that will be targeted for the conversion from RGB to HSV, then back to RGB and after conversion will be finished off by applying the colour correction to the right parts of the webpage. The adaption to code from algorithm was found and used with help from an earlier finding of RGBàHSV and HSVàRGB conversions (Jackson, 2013). The code was then adapted to work along with another function called changeNodesP (Figure 13) that advise which elements needed the colour correction. When the webpage was functioning and without bugs, all colour values were first corrected from hexadecimal colour values to RGB colour values (315fca5).

Figure 10 Example of RGB values in CSS file

After completing this, a colour conversion function was added to the webpage, that transforms

RGB colour space to HSV (Figure 11) colour space in order to render a more realistic colour

portrayal for the user who endures of colour vision deficiency. In the JavaScript function used

for this colour conversion that converts RGB colour space into HSV colour space the max

value is first retrieved from R, G or B as well as the minimum value. Then delta is calculated

by subtracting the minimum value from the maximum value. When this is determined, an if

statement is set by first comparing if the maximum value is equal to the minimum value then

the hue is set to 0. This is followed by a switch statement that executes different expression

depending on which match is made and the hue is divided by 6 and finally h, s and v are

returned.

Figure 11 JavaScript function of conversion from RGB to HSV

Next, the HSV values go back in a backward conversion to RGB in the set of 0 to 255 (Figure 12).

Figure 12 JavaScript function of conversion from HSV to RGB

A minor walkthrough with the changeNodesP function can tell that in order to do a conversion

of the all <p> elements on the webpage, the querySelectorAll() method is first applied for

gathering all the correct values. Regular Expressions (RegExp) method is then used for

extracting merely numbers when parsing the RGB-string. After parsing, the HSV objects are

retrieved from RGB values, this is followed by an if statement with a condition that says that

if the R value is greater than the B value, +0.7 will be added to the hue before the conversion

Figure 13 Function converting <p> tags on webpage

When the changeNodesP function proved to be successful with colour correcting all <p>

elements, the other text elements were included too, in this function. The other elements that were included were <h2>, <h1>, <a>, <button> and to two classes named .active and .divTable (Figure 14).

Figure 14 Selecting right elements for conversion

When all text elements performed the right actions, a new almost identical function to changeNodesP, called changesNodesB (Figure 15) was created. The changesNodesB function colour corrects the background colours on the webpage.

Figure 15 ChangesNodesB function

To activate the colour conversion, an onClick function was created with the id color-wheel.

When the user clicks the colour wheel found on the top of the page in the menu, the

changeNodesP and changesNodesB functions are activated (Figure 16).

Figure 16 Function activated when colour wheel is clicked

In order to determine whether the hypothesis, that the user will not be affected to the extent that they will abandon the webpage due to extended loading time, measurements had to be collected. In order to collect different measurements to evaluate, the webpage was tested with various conditions. All of these conditions were measured in milliseconds and the conditions that were produced was:

- Restaurant webpage without colour correction function;

- Restaurant webpage with colour correction function.

5.4 Pilot Study

The measurements that were collected for the pilot study are the response times of the original

webpage and then the colour corrected version. 1000 measurement values of each version were

collected and performed with a function called performance.now() that return page-loading

time in milliseconds and represent time as a floating-point number with up to a microsecond

precision (MDN web docs Mozilla, 2018). This indicates that the values collected will be with

a more preciseness which is of substance for providing good measurements. For the pilot study,

there has only been one element that has been passed through the conversion function and that

is the <p> tag element. They were saved and collected through LocalStorage in the web

browser. As can be seen in Figure 17 and 18, the milliseconds are presented in the y-axis and

number of measurements in the x-axis. The same structure follows in the second

measurements.

The noise of the webpage without colour correction (Figure 17) showed a few spikes that returned values that were 3 times as high as the remaining. The noise of the webpage with colour correction (Figure 18) showed a linear regression, which indicated that there was some kind of issue with the measurement test of page-loading time when the colour correction was activated. This linear regression might have been caused by a programming error, which was corrected for the second measurements.

Figure 18 The noise of webpage when activating colour correction

The average page-loading time, as previously mentioned, is measured in milliseconds. The baseline shows that it took in average 3341,81 milliseconds for the webpage to load, while the page-loading time with the applied colour correction took 7116,35 milliseconds (Figure 19).

The average of 3341,81 milliseconds would be acceptable if comparing to the response time

satisfaction to the user according to Hoxmeier et al. (2000), but questionable regarding the

page-loading time with colour correction, that had an average of 7116,35 milliseconds.

Figure 19 Average of page-loading in milliseconds

As evident in Figure 20 of the ANOVA Single Factor table, the P-value is 0, which indicates that the null hypothesis is rejected and shows that there is an affect between the baseline and colour correction. As can be seen from this pilot study, there was a significant difference between the baseline (original webpage) and colour correction.

Figure 20 ANOVA Single Factor of webpage with and without colour correction For the secondary measurements, more elements will be corrected, as for example background colour and other text colour elements on the webpage. There will also be corrections performed on the measurement function code for the second measurement procedure since there was an indication that some outcome proved be not be fully working while performing the pilot study.

Even though the results from the pilot study may not have been completely accurate since

Figure 18 showed a linear regression, there was still a significant difference from the baseline

which indicated an interest to continue with the measurement tests.

6 Evaluation

This section presents the results from the secondary measurements. As stated before, the goal with these measurements was to see if a real-time colour correction would be valuable to implement directly on a webpage, in order to provide a better web experience for those with CVD. Some changes were altered in the measurement method since the pilot study. To begin with, since Figure 18 indicated an error when collecting the values, it was decided to change the way of measuring the colour correction response time. Instead of measuring the complete webpage when activating the colour correction through the colour wheel in the menu, the values collected is solely the colour correction function loading time when activating the colour wheel on click function. The measurement values were also increased from 1000 to 5000 values. Furthermore, the secondary measurements also included:

- Background colour for <body> and all <div>’s with a distinctive colour set

- Text colour for <p>, headings and links and other values with distinctive colour sets The secondary measurements that have been collected are:

Measurement method W/wo colour conversion Additional information

5000 measurements of

page-loading in ms With Original webpage

5000 measurements of

page-loading in ms Without Additional extra data in

tables, to see difference with heavy data

5000 measurements of

page-loading in ms Without

All colour conversion code has been excluded from the webpage to see difference in

page-loading, no additional extra data

5000 measurements of extra page-loading time in

ms

With When colour conversion

button has been pressed

5000 measurements of extra page-loading time in

ms

With

When colour conversion button has been pressed with

extra data

Extra LOC With Extra LOC counted, that were needed for colour conversion

functions

6.1 Presentation

The experiment was performed on a single device and in the web browser Google Chrome, version 66.0.3359.139 (64 bit). The device is an Apple MacBook Pro (Retina, 13-inch, early 2015) with the following hardware specification:

Operating system: macOS Sierra, 10.13.3

CPU: 2.7GHz dual-core Intel Core i5 processor (Turbo

Boost up to 3.1GHz) with 3MB shared L3 cache

Memory: LPDDR3 SDRAM (8GB)

Graphics: Intel Iris Graphics 6100

Storage: 128GB SSD

6.2 Analysis

The highest spikes were observed in the range of 693 to 699 in the graph of the webpage with colour correction (Figure 21), this could be caused by e.g. the operating system prioritising other programs while running the tests or a network problem. Seeing as the highest spikes were six times as high as the mean, the probability for them to be caused by external reasons are most likely and therefore these spikes are excluded from the other charts. Just as Wohlin et al.

(2012) explains:

“It is important to remove data points which may make a completely valid relationship invalid, due to that, for example, an extreme outlier is included, which is not expected if replicating the study.”

Wohlin et al. (2012), pp. 172 6.2.1 Measuring page-loading time of webpage

The first part of the analysis compares the page-loading time of webpage with colour correction

script, webpage with extra data and colour correction script and of webpage without colour

correction script.

Figure 21 The noise of webpage with colour correction

A few, but very high spikes appeared on the webpage with colour correction and extra data (Figure 21), but not for the webpage without colour correction (Figure 22). There is no obvious reason why these spikes appeared for the colour correction webpages but not for the webpage without the colour correction script. Therefore, they are removed when performing the ANOVA tests where mean, standard error and confidence level are calculated.

Figure 22 The noise of webpage with colour correction and extra data

Figure 23 The noise of webpage without colour correction

The mean of the page-loading time for webpage both with and without colour correction proved to be quite equal, as can be seen in Figure 24. Also, the standard error bars showed a promising result with a low risk of repeating other measurements than the outcome of this study.

Figure 24 Average page-loading time of webpage with and without colour correction

After the highest spikes had been removed and an ANOVA Single Factor had been performed,

it showed that the webpage with colour correction had a small significant difference in contrary

to the webpage without colour correction. This concludes that the hypothesis is correct to that

the colour correction does not extend the page-loading time extensively, with the average page-

loading time for the webpage with colour correction being 3384,20 milliseconds and the

webpage without colour correction having an average of 3351,04 milliseconds. These average page-loading times would according to Hoxmeier et al. (2000) be acceptable to the user.

Figure 25 ANOVA Single Factor of webpage with and without colour correction The webpage with colour correction showed a greater variance than the webpage without colour correction. This can be caused, as earlier mentioned, by network errors or the operating system prioritizing other programs on the device. As can be seen from Figure 25, the P-value was below 0.05 and therefore the null hypothesis is rejected.

6.2.2 Measuring load time of colour correction script

The second part of the analysis compares loading time of colour correction on the original webpage and colour correction on webpage with extra data.

Figure 26 The noise of loading-time of colour correction function

The spikes that can be seen on the noise of loading-time for colour correcting functions (Figure

26) also provided some few and extensive spikes that were 4 to 8 times as high as the average.

Figure 27 Noise of loading-time of colour correction function with extra data It can be seen from the noise of colour correction function with extra data (Figure 27), that the highest spike was 20 times as high as the average.

Figure 28 Average loading-time with standard error for colour correction

function without and with extra data

As in the first part of secondary measurements, the highest spikes were removed in order to

present a more realistic average of the measurements. Even with a hundred extra tables

included on the webpage, the average loading-time was still under 1 millisecond. Though, it

should be taken into consideration that there is a natural delay as well for the colour correction

with extra data, due to the additional code.

Figure 29 Anova Single Factor of colour correction with and without extra data The null hypothesis is also rejected in the Anova Single Factor when comparing the colour correction function of the original webpage and the webpage with extra data since the P-value is 0.

6.2.3 Measurements of extra LOC required for colour correction

The extra LOC needed for the JavaScript colour correction function was 34 additional lines of code that converts the RGBàHSV, making a shift in the colour when and then back from HSVàRGB- Plus 1 additional line was needed in the HTML5 code for activating the script, which shows that there is not much extra effort needed to implement a colour correction function.

6.3 Conclusion

In the analysis, all the highest spikes of conducted measurements were removed, on account of that, the average was not representing the most frequent results. Also, these spikes can, as mentioned earlier, be caused by an e.g. network error or the operating system prioritising other programs while the tests were executed.

From the first part of the secondary measurements, the difference in response time between the

webpage with colour correction and the webpage without, were minor. Figure 24 showed that

the risk of receiving sample means that are deviating from the real mean is insignificant. The

results from the ANOVA Single Factor showed that the P-value was under 0.05, which rejected

the null hypothesis and there is no bias. The second part of the secondary measurements

revealed that both the colour correction function with and without extra data on the webpage

was executed with a good response time, both under 1 millisecond. Just as with the first part

of the secondary measurements, the P-value was also proved to be under 0.05 for the second

part.

7 Concluding remarks

7.1 Summary

The general goal of web accessibility is to give users with disabilities the same access to and use of information and that they should be able to use a system in the usual way as other users, without particular difficulties and without the help of others (Sik-Lányi et al., 2017).

For this project, a colour correction function was implemented on a webpage, for the purpose of exploring if it would affect the user’s experience in a negative way, measured in time. The intention with this study was to examine the possibility for a simpler solution for those with CVD, in order of making web content as accessible for them as for those who do not suffer from CVD.

The final remarks regarding implementing a real-time conversion are that the response time, as predicted, proved not to be excessive and the workload was not unreasonable. A colour correction directly in the web browser would therefore, be a good option for organizations, governments and other companies to use when developing web content.

7.2 Discussion

The advantages of implementing a CVD colour correction in the web browser are many. As mentioned earlier, the assistive technologies are most often solutions you have to activate by yourself (Schmitt et al., 2012) in your operating system or in downloaded assistive software.

A colour correction function with a noticeable button on the webpage would most likely make this assistive function distinct for the user and would be a valuable feature for those with CVD.

This colour correction function would even be of advantage for the groups of people with a non-diagnosed CVD (Jefferson & Harvey, 2007) since their colour related vision problems would not demand to install any extra software nor need to activate any other assistive technologies.

The alternative hypothesis was that the conversion would not elongate the page-loading time when activating the colour correction. In the analysis, the results proved this to be right. As stated before, there was a change in the secondary measurement test due to results from the first measurements where there seem to be a programming error in the code for the function of measuring the colour correction. For the secondary measurement tests, these errors were corrected and the results showed no signs of faults when executing the measurement scripts.

The conversion script could be improved to be shorter in LOC, cleaner and more effective.

However, even though this is the first version of a real-time colour correction with JavaScript, this version still proved to be sufficient towards the goal of creating a simple colour correction for those with red-green CVD.

The extra LOC needed for the colour conversion, as mentioned before, was not excessive.

Proving that this type of additional function on a website would not exceed the workload for

the developer and would also extend the target audience. Many developers have according to

Kowto (2012) shown a great interest in making the web more accessible if they had better

knowledge of how to implement these actions. As Kowtko (2012) also stated, there are over

80 million people with different types of disabilities only within the EU member states and

about one-quarter of these suffer from some kind of colour vision deficiency. Hence, it would

be of great interest to execute further studies to see if this type of colour correction would

benefit users with CVD. Another interesting aspect to measure would be to conduct a survey with colour tests between RGB and HSV, to see if this does provide a great difference for the human eye interpretation, as Saravanan et al. (2016) has stated.

7.3 Future work

For future work, it would be interesting to test this recolouring process on individuals with colour vision deficiency, to see if this idea would contribute to a better web experience.

Questions that could be answered from e.g. a survey, could be to see if this recolouring process fulfil its function, i.e. accomplishing a better interpretable colour set for those with red-green CVD to be able to extract important information, that users without CVD not might be troubled with.

Another interesting aspect would be to conduct tests to see how long time a CVD user would be willing to wait for a colour converted website, and for that produce a heavier loaded page, to receive more knowledge about how valuable this kind of function would be, when accessing web content. Needless to say, the point is not to test people’s patience, but as Ruminski et al.

(2010) declared, it has shown that high-resolution images take a longer time to translate since there are more pixels to translate. The article from Ruminski et al. (2010) shares several similarities to this study since they also collected response time of colour transformation and too adopted a quantitative experimental method. But as mentioned before, their focus was on image processing method and was not executed with the same programming language as in this study. Therefore, it would be of great interest to perform measurement tests on colour corrected webpages with additional colour transformed images to see how this affects the response times.

It would also be of great interest to run these measurements on other operating systems or even

other browsers, to see if the same spikes would appear, as in the measurements performed in

this study. This to evaluate why the spikes in Figure 21 were so exaggerated from the mean

and not for Figure 22. Lastly, it would be profitable to explore other colour correction options

to include several CVD conditions altogether and not solely anomalous trichromats as in this

study, where the target was those with red-green CVD. This with the purpose of creating web

content that is available to as many individuals as possible, without the need for any additional

assistive technologies.

References

AChecker. (2011) Web Accessibility Checker. Available at:

https://achecker.ca/checker/index.php [Accessed 2018-03-20].

Chernov, V., Alander., J & Bochko, V. (2015) Integer-based accurate conversion between RGB and HSV color spaces. Computers & Electrical Engineering 46. Elsevier. pp. 328-337.

Ching, S-L. & Sabudin, M. (2010) Webpage image colour transformation for the colour blind. 2010 2nd International Conference on Computer Technology and Development.

IEEE. pp. 255-259.

European Union. (2017) Web Accessibility. Available at: https://ec.europa.eu/digital-single- market/en/web-accessibility [Accessed 2018-01-02].

Flanagan, D. (2011) JavaScript: The Definitive Guide, Sixth Edition. O’Reilly Media, Inc.

Flatla, D. R. & Gutwin, C. (2012) "So that's what you see": building understanding with personalized simulations of colour vision deficiency. In Proceedings of the 14th international ACM SIGACCESS conference on Computers and accessibility (ASSETS '12).

ACM. pp. 167–174.

Gilbertson, T. D. & Machin C H. C. (2012) Guidelines, icons and marketable skills: an accessibility evaluation of 100 web development company homepages. In Proceedings of the International Cross-Disciplinary Conference on Web Accessibility (W4A '12). ACM.

Article 17, 4 pages.

Hoxmeier, J. A. & DiCesare, C. (2000) System Response Time and User Satisfaction: An Experimental Study of Browser-based Applications. In AMCIS 2000 Proceedings 347. pp.

140-145.

Huang, Y-H., Wang, P-C., Chou, CC, & Lin, K-P. (2011) An automatic selective color transfer algorithm for images. In Proceedings of the 2011 ACM Symposium on Applied Computing (SAC '11). ACM. pp. 66-71.

Jackson, M. (2013) RGB, HSV, and HSL color conversion algorithms in JavaScript. GitHub Gist. Available at: https://gist.github.com/mjackson/5311256/revisions. [Accessed 2018- 04-16].

Jefferson, L. & Harvey, R. (2006) Accommodating color blind computer users. In Proceedings of the 8th international ACM SIGACCESS conference on Computers and accessibility (Assets '06). ACM. pp. 40–47.

Jefferson, L. & Harvey, R. (2007) An interface to support color blind computer users.

In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '07). ACM. pp 1535-1538.

Jorde, L. B., Carey, J. C. & Bamshad, M. J. (2015) Medical Genetics. Elsevier. pp. 81,86.

Kowto, M. (2012) Using assistive technologies to improve lives of older adults and people with

disabilities. In Systems, Applications and Technology Conference (LISAT). IEEE. pp. 1-6.